Deep Learning Assisted Design of EM Skin Meta-Atoms within the System-by-Design

An innovative System-by-Design (SbD) method is proposed for the design of non-parametric electromagnetic skin (EMS) meta-atoms. The developed technique leverages a smart encoding of the unknowns allowing to exploit physics knowledge and to enable a favorable environment for the optimization-driven solution of the arising synthesis problem. This latter is efficiently solved by means of a customized SbD implementation relying on a Deep-Learning (DL)-based Surrogate Model (SM) of the pixel-based reflective meta-atom and the exploration of the solution space by means of an Integer-Coding Genetic Algorithm (ICGA) strategy. A preliminary proof-of-concept is shown to assess the capabilities of the proposed method.